Meta's New 'State-Of-The-Art' AI Image Generator Could Give DALL-E A Run For Its Money
AI efforts from Google and OpenAI have garnered a lot of attention in recent months, but Meta believes its new model could be one of the best available — at least when it comes to AI art generation, anyway. The company has just launched "CM3leon," which is pronounced "chameleon," an image generator that can transform text-prompts into pictures and vice-versa. According to a statement on the Meta website, CM3leon is "the first multimodal model trained with a recipe adapted from text-only language models." Because of this, Meta says the model achieves "state-of-the-art performance for text-to-image generation, despite being trained with five times less compute than previous transformer-based methods."
In simple terms, you can use the new tool to create new, complex, images from scratch or make edits to existing images. Edits can include turning summer into winter, changing the color of the sky, or anything else you can summarize with a prompt. As far as generated images go, the limit is your imagination. Users can also add a picture and ask the AI model questions about it, whether that's an inquiry about specific aspects of the photo, or even asking the model to create a detailed caption of the image in question. Other supported features include object-to-image, segmentation-to-image, and the addition of a "super resolution" stage that bumps up the resolution of created pictures.
There is a lot of competition
Meta's image generator is certainly unique, and does things very well, but it is far from the only game in town. Similar tools have been around for a while now, and most major players have one in some form. OpenAI's DALL-E hit the headlines last year, appears in several forms, and has a strong following. The company's GPT-4 model is also capable of image generation, though a plugin is needed if you're using OpenAI's web tool. Adobe was similarly quick to add AI to its suite of services. "Firefly," the company's generative AI model, is available to use for free — though subscribers get access to Adobe's stock photo library and thus a more effective version of the AI tool that relies on it. Google Bard also uses Firefly for image generation purposes. Then you have powerful tools that require a bit of work to get them going, and decent prompts if you wish to use them effectively. This category includes the likes of Stable Diffusion and Midjourney.
As for when you can try Meta's new tool, we don't know. The company hasn't announced when, or if, it will receive a public release. Meta certainly has enough going on already with the launch of a new social media platform, Threads, and the steady collapse of "the Metaverse" to contend with, but we're hoping this new AI image generator gets a public release in the weeks to come.